[Bf-cycles] Fwd: Re: Blender Viewport performance issue after Cycles render (since changing to CUDA Toolkit 5.0)

David Black db4tech at yahoo.co.uk
Mon Jun 24 00:56:36 CEST 2013


After further tests, systematically stripping my complex scene, I have 
been able to narrow it down to which object is affected. Leaving just a 
heavy particle carpet (particles made from complex geometry as an Object 
Group).

I have also been able to identify when the Viewport F12 Cycles render 
issue happens.

With the particle carpet, (Particle settings) if render Emitter and 
Parents are checked, Cycles Viewport performance is okay after Cycles 
render, if only Parents is checked, then the Viewport slows right down 
to a crawl until, either Viewport shading is changed to Texture shading 
then back to Solid shading, or scene is reloaded.

Hope this is helpful!
David

-------- Original Message --------
Subject: 	Re: [Bf-cycles] Blender Viewport performance issue after 
Cycles render (since changing to CUDA Toolkit 5.0)
Date: 	Sun, 23 Jun 2013 22:17:12 +0100
From: 	David Black <db4tech at yahoo.co.uk>
Reply-To: 	bf-cycles at blender.org
To: 	bf-cycles at blender.org



You are welcome.

* What is the performance like with the NVidia card, is it also 2fps
or somewhere in between?


Tested NVIDIA GTX 580m (cannot test GTX 580 yet due to data backup), 
same result, Viewport slowdown to a crawl after F12 Cycles render. Also 
tried an Intel HD 3000, plus a virtual machine, VMware running Windows 7 
x64, same result in each case.

As an aside, your requested tests have produced an interesting (none 
Cycles) finding, as I expected the GTX 580m is slower than a HD 3000 for 
Solid shaded Viewport (due to Double sided Normals), what I find 
interesting is, Texture shaded Viewport is faster (smoother) than Solid 
with a GTX 580m. If okay to ask, is this due to present OpenGL 
implementation, or does Texture shaded Viewport automatically disable 
doubled sided shading?

Texture shaded is a little slower than Solid for HD 3000, which I 
imagine is due to processing performance.

Would you still like me to test Viewport performance after F12 Cycles 
render with a GTX 580, once data backup is finished?

David
-- 

3d-designs-davidblack.blogspot.com 
<http://www.3d-designs-davidblack.blogspot.com>

On 23/06/2013 20:12, Brecht Van Lommel wrote:
> Thanks for the detailed reported, it will help a lot in fixing this bug.
>
> On Sun, Jun 23, 2013 at 8:00 PM, David Black<db4tech at yahoo.co.uk>  wrote:
>> On 23/06/2013 18:03, Brecht Van Lommel wrote:
>>
>> I have no idea how this is possible, all the toolkit changes is code
>> that runs on the NVidia graphics card, not how it interacts with other
>> graphics drivers. So perhaps it's another recent change in Cycles or
>> elsewhere in Blender that is causing this?
>>
>> Apologies, not Toolkit 5.0 related, after performing your tests below
>> discovered it goes back further.
>>
>>
>> * What is the version number of the build before the toolkit change
>> that still gave good performance (to narrow down the revisions)?
>>
>> blender-2.67-r57293-win64 is okay, Viewport performance same before and
>> after F12 Cycles render.
>>
>> blender-2.67-r57348-mingw64 Is okay except for border render issue,
>> otherwise same as above.
>>
>> blender-2.67-r57476-win64 First versions (I have) that shows the issue,
>> Viewport performance same as previous version until F12 Cycles render, then
>> very slow jerky Viewport performance.
>>
>>
>> * Does this really only happen with GPU rendering, CPU rendering does
>> not show the same issue?
>>
>> Interesting test idea. Same issue after either GPU or CPU F12 Cycles render
>>
>>
>>
>> * What is performance like with VBO's switched off before rendering
>> (in case the rendering somehow switches them off)?
>>
>> Not as smooth, around 10-15fps but still faster than after rendering, which
>> then feels around 2fps.
>>
>>
>>
>> * What is the performance like with the NVidia card, is it also 2fps
>> or somewhere in between?
>>
>> Not yet tested, once wiring configurations are changed will email results.
>> Thought quick reply with above answers would be appreciated.
>>
>>
>> * Is a Blender restart needed or does reopening the scene solve the issue
>> too?
>>
>>
>> Interesting, reopening the same scene without restarting Blender also solves
>> the issue, until an F12 Cycles render is performed.
>>
>> David
>>
>>
>> Can anyone else confirm this problem?
>>
>> On Sun, Jun 23, 2013 at 5:36 PM, David Black<db4tech at yahoo.co.uk>  wrote:
>>
>> Hi Brecht,
>>
>> Just discovered an interesting issue, seems to have been introduced since
>> Blender changed to CUDA Toolkit 5.0. Not related to Cycles render speeds,
>> which consequently have received a great boost after code re-factoring.
>> Issue is connected with Viewport performance after performing a Cycles
>> render.
>>
>> Issue outlined below:
>>
>> Upon opening a complex scene, with a heavy particle system (particles
>> Viewport setting of 1, Child particles 0), Viewport navigation (VOBs
>> enabled) with an Intel HD 4000 (part of i7 CPU) is extremely fluid, like
>> 30fps or more, (much slower on GTX 580 due to double sided Normals
>> restrictions).
>> With recent Blender versions, everything is fine until a F12 Cycles render
>> is performed (does not happen with Viewport render), after which Viewport
>> navigation becomes extremely sluggish, feels like 2fps. Hiding the heavy
>> particle object returns full Viewport performance, once object is unhidden
>> Viewport is back to a crawl. Closing and opening Blender with the same scene
>> and everything is back to normal, until the next F12 render is performed.
>>
>> Blender versions prior to Toolkit 5.0 changes, Viewport performance remains
>> just as fluid after a F12 Cycles render, also tested Blender 2.67b.
>>
>> Today's BuildBot version (official) blender-2.67-r57660-win64 produces the
>> sluggish Viewport issue after a F12 Cycles render.
>>
>> David
>> --
>>
>> 3d-designs-davidblack.blogspot.com
>>
>>
>> _______________________________________________
>> Bf-cycles mailing list
>> Bf-cycles at blender.org
>> http://lists.blender.org/mailman/listinfo/bf-cycles
>>
>> _______________________________________________
>> Bf-cycles mailing list
>> Bf-cycles at blender.org
>> http://lists.blender.org/mailman/listinfo/bf-cycles
>>
>>
>>
>> _______________________________________________
>> Bf-cycles mailing list
>> Bf-cycles at blender.org
>> http://lists.blender.org/mailman/listinfo/bf-cycles
>>
> _______________________________________________
> Bf-cycles mailing list
> Bf-cycles at blender.org
> http://lists.blender.org/mailman/listinfo/bf-cycles
>



-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.blender.org/pipermail/bf-cycles/attachments/20130623/66c0f5d1/attachment.htm 


More information about the Bf-cycles mailing list